mersenneforum.org > Data Dual core anyone?
 Register FAQ Search Today's Posts Mark Forums Read

2006-08-25, 22:27   #12
drew

Jun 2005

2×191 Posts

Quote:
 Originally Posted by Prime95 Are they really stuck or are they making progress toward completing stage 2? I'm changing the "Using nMB of memory." output to "Using nMB of memory. Processing x relative primes (y of z already processed)." This should make it clear that forward progress is being made.
I had stopped it because it looked like it was in an infinite loop, but now that think it might still be working, it does appear to be making progress. It appears to run a test for a couple of minutes before moving on to the next one, however, which makes me suspicious, because I'm used to a test taking a day or two to complete P-1 factoring. However, that's in the 35 million range and these are in the 28 million range (and using different assumptions having been LL tested once already.

I wish I could copy and paste the text output so I could send it to you.

Here's what's in results.txt so far:

[Fri Aug 25 15:15:05 2006]
UID: admin_user_anon/M90, M28781689 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
UID: admin_user_anon/M90, M28515913 completed P-1, B1=150000, B2=2925000, Wc1: 88888888
UID: admin_user_anon/M90, M28524121 completed P-1, B1=150000, B2=2925000, Wc1: 88888888
UID: admin_user_anon/M90, M28786517 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
UID: admin_user_anon/M90, M28801057 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
UID: admin_user_anon/M90, M28527281 completed P-1, B1=150000, B2=2925000, Wc1: 88888888
UID: admin_user_anon/M90, M28801693 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
UID: admin_user_anon/M90, M28537589 completed P-1, B1=150000, B2=2925000, Wc1: 88888888
UID: admin_user_anon/M90, M28806137 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
UID: admin_user_anon/M90, M28551451 completed P-1, B1=150000, B2=2925000, Wc1: 88888888
[Fri Aug 25 15:20:16 2006]
UID: admin_user_anon/M90, M28807003 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
UID: admin_user_anon/M90, M28556123 completed P-1, B1=150000, B2=2925000, Wc1: 88888888
UID: admin_user_anon/M90, M28559429 completed P-1, B1=150000, B2=2925000, Wc1: 88888888
UID: admin_user_anon/M90, M28810057 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
UID: admin_user_anon/M90, M28559789 completed P-1, B1=150000, B2=2925000, Wc1: 88888888
UID: admin_user_anon/M90, M28815713 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
UID: admin_user_anon/M90, M28564187 completed P-1, B1=150000, B2=2925000, Wc1: 88888888
UID: admin_user_anon/M90, M28825669 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
UID: admin_user_anon/M90, M28571113 completed P-1, B1=150000, B2=2925000, Wc1: 88888888
UID: admin_user_anon/M90, M28826041 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
UID: admin_user_anon/M90, M28594309 completed P-1, B1=150000, B2=2925000, Wc1: 88888888
UID: admin_user_anon/M90, M28826047 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
[Fri Aug 25 15:25:23 2006]
UID: admin_user_anon/M90, M28826449 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
UID: admin_user_anon/M90, M28618111 completed P-1, B1=155000, B2=2945000, Wc1: 88888888
UID: admin_user_anon/M90, M28828421 completed P-1, B1=155000, B2=2945000, Wc1: 88888888

Memory usage in task manager appears to jump around erratically.

Drew

2006-08-25, 23:16   #13
Prime95
P90 years forever!

Aug 2002
Yeehaw, FL

22·1,873 Posts

Quote:
 Originally Posted by drew Wc1: 88888888
This indicates you've got the special "server testing" version which bypasses all the real work (to pummel server with lots of requests and results).

I thought I had uploaded the proper build. Apparently I goofed. I'll upload another version. Sorry, welcome to the bleeding edge!

get ftp://mersenne.org/gimps/p95tst2.zip

Last fiddled with by Prime95 on 2006-08-25 at 23:23

2006-08-25, 23:29   #14
drew

Jun 2005

2×191 Posts

Quote:
 Originally Posted by Prime95 I thought I had uploaded the proper build. Apparently I goofed. I'll upload another version. Sorry, welcome to the bleeding edge!
I work in an R&D department, so I know what it's like.

Thanks for the update.

Drew

2006-08-26, 02:34   #15
James Heinrich

"James Heinrich"
May 2004
ex-Northern Ontario

3,361 Posts

I'm trying it out for a few exponents. Some initial observations:

1) My Opteron is dual-core, but the cores are not hyperthreaded... (see attachment)

2) When doing the initial self-test, might it not be a good idea to run test concurrently on all cores, to ensure that any heat issues brought about by full-load running would appear during the test, not just during actual work? My CPU is running 5-6C hotter during both-cores running compared to a single core during the test.

3) It might be a good idea to do a child-window re-tile when un-tray'ing the window (that phrase was over-hyphenated! ) That is, when I initially open the main window the child windows are tiled. If I maximize the main window the child windows are nicely automatically re-tiled, but when I minimize to tray and then restore, the parent window comes up non-maximized but the child windows are still set for the full-screen tiled layout. (if that paragraph made any sense )

4) I agree with drew's post above that Prime95 should try and do some intelligent stage1/stage2 management, that it should try and run stage1 on one core while stage2 is running (for a different exponent, obviously) on the other core. Future expansion to 4 (or more) cores will obviously lead to inevitable competition for memory, but with only 2 threads it should be possible to (mostly) avoid running stage2 with only half the available memory, at least most of the time. It does require some more work on your (Prime95) part to write code to do the intelligent work scheduling, but it should be reasonably easy, no?

5) I'm not sure if this is just random, but the first 2 attempts at running the self-test failed (on 8th and 9th tests, respectively). I realize I'm running this CPU overclocked, but I've had it clocked 4% faster (2.6GHz, compared to 2.5GHz now) for the last 3 months, running Prime95 (mostly trial factoring) on both cores with no problems in LL self-test or anything else. The third attempt at the (1536K) self-test ran through 20 tests just fine, I'm not sure why...

6) It would be nice to be able to select / copy text from the child windows

7) The benchmark page claims that 1536K is for 24.93M-29.69M, and yet the self test runs from 29.3M down to 20.?M -- why does it run to such a low level if it's not going to be used in practice?

9)
Quote:
 This ought to work: Pfactor=exp,how_far_factored,double_check_flag [thread #2] Pfactor=...
OK, that works, but it would be much nicer if Prime95 itself could pull from a single worktodo list and assign stage1 and stage2 work to different threads as needed (see #4 above)
Attached Thumbnails

 2006-08-26, 02:43 #16 James Heinrich     "James Heinrich" May 2004 ex-Northern Ontario 3,361 Posts In the benchmark I encounterered a strange problem on 65-bits. In the thread window it showed "Unknown", in results.txt it shows "0.000 ms.": Code: Best time for 58 bit trial factors: 4.670 ms. Best time for 59 bit trial factors: 4.679 ms. Best time for 60 bit trial factors: 4.661 ms. Best time for 61 bit trial factors: 4.682 ms. Best time for 62 bit trial factors: 8.495 ms. Best time for 63 bit trial factors: 8.496 ms. Best time for 64 bit trial factors: 11.082 ms. Best time for 65 bit trial factors: 0.000 ms. Best time for 66 bit trial factors: 11.014 ms. Best time for 67 bit trial factors: 10.970 ms. Also, from this thread, I'll repeat my request: I notice the benchmark goes up to much larger FFT sizes now, but it still only includes trial-factoring times up to 67 bits, not the 71 bits that would be used on ~70M+. Maybe even benchmarking up to 74 bits would be nice.
 2006-08-26, 02:44 #17 drew     Jun 2005 2×191 Posts FWIW, when I ran it, it ran the self-test concurrently and successfully on both threads. I think I prefer having the work for two threads separated like they are in the worktodo file. Drew
 2006-08-26, 12:57 #18 drew     Jun 2005 2·191 Posts A couple of more observations, for what they're worth: It appears to have allocated slightly more RAM to the smaller exponent. I've allocated 1024 MB for Prime95 (half of my total RAM) Exponent M28515913 is using B1=150000, B2=2925000, 503 MB RAM Exponent M28781689 is using B1=155000, B2=2945000, 491 MB RAM M28515913 begain stage 2 a couple of minutes before the larger exponent. Thread 1 ran for a couple of minutes using the full allotted memory. When thread 2 began, it reduced its memory by half, then re-displayed the B1, B2 and odds of finding a factor values (which amount memory is assumed when determining these values?) It's behaving nicely. Using 1054,364K in Task Manager, and the whole system is using 1.34 GB, so it's not thrashing or anything. Drew
2006-08-26, 13:46   #19
Prime95
P90 years forever!

Aug 2002
Yeehaw, FL

1D4416 Posts

Quote:
 Originally Posted by James Heinrich ...but it still only includes trial-factoring times up to 67 bits, not the 71 bits that would be used on ~70M+. Maybe even benchmarking up to 74 bits would be nice.
Benchmark times for 66 bits and higher should be the same.

2006-08-26, 13:47   #20
Prime95
P90 years forever!

Aug 2002
Yeehaw, FL

11101010001002 Posts

Quote:
 Originally Posted by James Heinrich My Opteron is dual-core, but the cores are not hyperthreaded...
Figures. I'm using code straight from Intel's documentation.

2006-08-26, 13:53   #21
Prime95
P90 years forever!

Aug 2002
Yeehaw, FL

22×1,873 Posts

Quote:
 Originally Posted by James Heinrich 2) When doing the initial self-test, might it not be a good idea to run test concurrently on all cores, to ensure that any heat issues brought about by full-load running would appear during the test, not just during actual work? My CPU is running 5-6C hotter during both-cores running compared to a single core during the test.
I've actually thought about eliminating the one-hour self-test. I'm not convinced it keeps bad machines out of GIMPS as they eventually pass the one-hour test and start crunching.

As to your point, in a normal installation you'd get LL tests for both threads. Both would start the self-test at about the same time, so yes all cores would be self-testing at the same time.

The torture test dialog box now lets you choose how many threads to run.

Last fiddled with by Prime95 on 2006-08-26 at 18:44

2006-08-26, 14:01   #22
Prime95
P90 years forever!

Aug 2002
Yeehaw, FL

22×1,873 Posts

Quote:
 Originally Posted by James Heinrich 3) It might be a good idea to do a child-window re-tile when un-tray'ing the window (that phrase was over-hyphenated! ) That is, when I initially open the main window the child windows are tiled. If I maximize the main window the child windows are nicely automatically re-tiled, but when I minimize to tray and then restore, the parent window comes up non-maximized but the child windows are still set for the full-screen tiled layout. (if that paragraph made any sense ) 6) It would be nice to be able to select / copy text from the child windows
The tiling/scrolling code does need work. Maybe some MFC experts could help out...

Quote:
 4) I agree with drew's post above that Prime95 should try and do some intelligent stage1/stage2 management,
I'll see if I can come up with something better.

Quote:
I've thought about that. I'm not sure if that's a useful feature in everyday usage.

Quote:
 9) OK, that works, but it would be much nicer if Prime95 itself could pull from a single worktodo list and assign stage1 and stage2 work to different threads as needed (see #4 above)
Again, this feature may require more effort than it is worth. Certainly a PrimeNet user would have no use for this feature. The current scheme is no different than you presently have running two copies of prime95 v24 on a dual core machine.

 Similar Threads Thread Thread Starter Forum Replies Last Post Rodrigo PrimeNet 4 2011-07-30 14:43 Rodrigo Hardware 6 2010-11-29 18:48 xorbe PrimeNet 4 2009-04-04 15:32 patrik Hardware 3 2007-01-07 09:26 R.D. Silverman Hardware 12 2005-02-20 21:46

All times are UTC. The time now is 07:43.

Sat May 15 07:43:02 UTC 2021 up 37 days, 2:23, 0 users, load averages: 2.73, 2.14, 2.14

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.